Capital structure has been a topic of significant interest since the beginnings of academic finance. Numerous recent empirical contributions would be appropriate as for us to set forward as the benchmark model. We base the leverage results below on specifications from Lemmon, Roberts, and Zender (2008), the reason being in part that they also suggest that firm fixed effects are important, particularly insofar as the fixed
effects capture unobserved time-invariant factors that generate surprisingly stable capital structures.
9.1 Book leverage
The benchmark specification is a pooled OLS regression, without firm or manager fixed effects, which is based on the explanatory variables in Lemmon, Roberts, and Zender (2008, Table II, Panel A, Column 3), plus selected observable managerial attributes. The dependent variable is total debt scaled by firm assets. Benchmark adjusted R2 is 0.23 (Model (1), Panel A, Table 12). Adding firm fixed effects increases adjusted R2 to 0.67 (Model (2)), while manager fixed effects boosts adjusted R2 to 0.71 (Model (3)). Model (4) includes both firm and manager fixed effects to yield adjusted R2 = 0.74.
As Panel B of Table 12 indicates, when normalized by variation of book leverage explained by the model, observable firm attributes, observable manager attributes, firm fixed effects, manager fixed effects, and year fixed effects, contribute 10.8%, 0.7%, 62.5%, 25.6%, and 0.5% of model R-squared. Column F1 in Figure 1 represents graphically these components of variation.
These results are consistent with the results and conclusions of Lemmon, Roberts, and Zender (2008). They find that corporate capital structures are stable over long periods of time and that this feature of the leverage data-generating process is present after controlling for firm entry and exit and for previously identified determinants of capital structure. Overall, they show that much of the variation in capital structure is time-invariant and that much of that variation is not explained by existing empirical specifications. This is consistent with Parsons and Titman (2008) and Graham and Leary
(2011) who suggest more work is needed to identify what is captured in firm fixed effects but are missing from our models.
Compare Models (1) and (4) to discern the primary changes in empirical results from including manager and firm fixed effects. The coefficients on log (sales) and the dividend payer dummy (and manager tenure) change from positive (negative) and significant to negligible and insignificant.
9.2 Market leverage
The dependent variable is total debt scaled by the sum of total debt and market equity. The baseline specification is a pooled OLS regression, without firm or manager fixed effects, which is based on the explanatory variables in Lemmon, Roberts, and Zender (2008, Table II, Panel A, Column 6), plus selected observable managerial attributes. Benchmark adjusted R2 is 0.39 (Model (1), Panel A, Table 13). Adding firm fixed effects increases adjusted R2 to 0.77, while manager fixed effects boosts adjusted R2 to 0.76. Model (4) includes both firm and manager fixed effects to yield adjusted R2 = 0.79.
Panel B of Table 13 indicates that, when normalized by variation of market leverage explained by the model, observable firm attributes, observable manager attributes, firm fixed effects, manager fixed effects, and year fixed effects, contribute 22.5%, 0.1%, 51.6%, 24.6%, and 1.2% of model R-squared. Column F2 in Figure 1 represents graphically these components of variation.
These results and those for book leverage are consistent with the strong explanatory power for firm fixed effects reported in Lemmon, Roberts, and Zender (2008). Nonetheless, managerial fixed effects also represent a significant component of variation in both book and market leverage explained, which is consistent with the notion
that managerial characteristics, such as risk aversion, affect corporate financial policy (see Cronqvist, Makhija, and Yonker, 2010). Moreover, overconfident managers exhibit strong debt conservatism and pecking-order preferences (Malmendier, Tate, and Yan, 2011). Furthermore, Berger, Ofek, and Yermack (1997) show managers who have more control would keep leverage ratio low. Other important factors are likely to be cost of borrowing as determined by credit risk, relationships with lenders, and the implicit interest rate. In addition, firm culture will keep leverage and cash holding persistent over time (Cronqvist, Low, and Nilsson, 2007).
Comparing Models (1) and (4), the primary changes in empirical results from including manager and firm fixed effects are that the dividend payer dummy changes from small and insignificant to positive and insignificant, while significant coefficients on manager tenure, the CEO indicator, and the board seat indicator become small and insignificant.
9.3 Cash holdings
The baseline specification is a pooled OLS regression, without firm or manager fixed effects, which is based on the explanatory variables in Harford, Mansi, and Maxwell (2008, Table 3, Column 1), plus selected observable managerial attributes. The natural logarithm of the cash to sales ratio is the dependent variable. Benchmark adjusted R2 is fairly large at 0.41 (Model (1), Panel A, Table 14). Adding firm fixed effects yields adjusted R2 of 0.80, manager fixed effects alone delivers adjusted R2 of 0.80, and using both gives adjusted R2 = 0.82.
Panel B of Table 14 indicates that observable firm attributes, observable manager attributes, firm fixed effects, manager fixed effects, and year fixed effects, contribute
17.0%, 0.80%, 51.0%, 28.8%, and 2.4% of model R-squared.6 Column F3 in Figure 1
represents graphically these components of normalized variation. Like book and market leverage, firm observables and fixed effects dominate manager observable and fixed effects.
Frank and Goyal (2009) show industry median leverage is an important correlate, acting as a proxy for company specific intangibility, regulation, stock variance, and uniqueness. Obviously, this proxy is so noisy that the firm fixed effects may capture a large portion of these factors.
Comparing Models (1) and (4), including manager and firm fixed effects alters parameter estimates and inference in numerous cases. Significant coefficients in Model (1) that are insignificant in Model (4) with both fixed effects are the director indicator, G index, cash flow volatility, and market-to-book. Variables that become significant are firm size (-), R&D intensity (-), and manager age (-). Omitted variables and other endogeneity problems may be a significant difficulty in the analysis of cash holdings.